# Multi-hop Question Answering

**Authors:** Vaibhav Mavi, Anubhav Jangra, Adam Jatowt  
**arXiv:** [2204.09140](https://arxiv.org/abs/2204.09140)  
**Date:** 2022-04-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2204.09140v2/

## Key points

- Question Answering (QA) is a significant research area for language understanding and knowledge retrieval tasks.
- Multi-Hop QA (MHQA) has become a prominent focus, involving answering questions that require multiple steps of reasoning.
- MHQA enhances Natural Language Processing (NLP) systems by combining and extracting information from various sources.
- Challenges in MHQA include diverse tasks requiring multi-hop reasoning, leading to variations in datasets and models.
- A new book aims to define the MHQA task, organize existing frameworks, and provide best practices for constructing MHQA datasets.

## Summary

, , , , 

The task of Question Answering (QA) has been a subject of significant research interest due to its relevance in language understanding and knowledge retrieval tasks. Recent advancements in simple QA tasks have led to a shift towards more complex settings, with Multi-Hop QA (MHQA) emerging as one of the most researched tasks in recent years. MHQA involves answering natural language questions that require extracting and combining multiple pieces of information through multiple steps of reasoning. This question necessitates gathering two pieces of information: identifying the record holder for Argentine PGA Championship tournaments and determining the number of tournaments they have won. The ability to answer multi-hop questions and engage in multi-step reasoning can significantly enhance the effectiveness of Natural Language Processing (NLP) systems. The field of MHQA has experienced growth with the development of high-quality datasets, models, and evaluation strategies. However, the abstract nature of 'multiple hops' has resulted in a diverse range of tasks requiring multi-hop reasoning, leading to variations in datasets and models that pose challenges for generalization and surveying within the field. In response to these challenges, this book aims to provide a comprehensive definition of the MHQA task, organize existing MHQA frameworks, and outline best practices for constructing MHQA datasets. By offering a systematic introduction and structuring existing attempts at tackling this complex task, this book serves as a valuable resource for researchers seeking to delve into the intricacies of Multi-Hop Question Answering.